• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Is negative feedback fully effective in case of complex music waveforms?

Gorgonzola

Major Contributor
Forum Donor
Joined
Jan 27, 2021
Messages
1,017
Likes
1,398
Location
Southern Ontario
If heard it assert the the efficacy of negative feedback, and validity of distortion measurements, is really only effective in the case of steady-state signals. Is this true?

It seems to me that there is some latency in a feedback loop, so on the face of it, it sounds reasonable that the feedback might be too slow in case of complex music material. Could this be true? Also most seems it would be true if feedback took milliseconds to arrive but not if it took nanoseconds. I even heard is assert than on account of too-slow response, the feedback in case of complex forms is trying to correct as situation that is already in the past tense. :oops:

Does any of this make sense?
 

Killingbeans

Major Contributor
Joined
Oct 23, 2018
Messages
4,088
Likes
7,544
Location
Bjerringbro, Denmark.
 

antennaguru

Senior Member
Joined
Jun 16, 2021
Messages
391
Likes
416
Location
USA
If heard it assert the the efficacy of negative feedback, and validity of distortion measurements, is really only effective in the case of steady-state signals. Is this true?

It seems to me that there is some latency in a feedback loop, so on the face of it, it sounds reasonable that the feedback might be too slow in case of complex music material. Could this be true? Also most seems it would be true if feedback took milliseconds to arrive but not if it took nanoseconds. I even heard is assert than on account of too-slow response, the feedback in case of complex forms is trying to correct as situation that is already in the past tense. :oops:

Does any of this make sense?
Yes, excessive negative feedback can become destructive when it tries to pass a sharp impulse beyond the normal headroom of the system, and then the latency detracts from the recording. This is especially obvious when comparing phono stages with a record that has a pop/tick - which can linger much longer audibly versus quickly dissipating such that less attention is drawn to it.
 

Doodski

Grand Contributor
Forum Donor
Joined
Dec 9, 2019
Messages
20,745
Likes
20,757
Location
Canada
If heard it assert the the efficacy of negative feedback, and validity of distortion measurements, is really only effective in the case of steady-state signals. Is this true?

It seems to me that there is some latency in a feedback loop, so on the face of it, it sounds reasonable that the feedback might be too slow in case of complex music material. Could this be true? Also most seems it would be true if feedback took milliseconds to arrive but not if it took nanoseconds. I even heard is assert than on account of too-slow response, the feedback in case of complex forms is trying to correct as situation that is already in the past tense. :oops:

Does any of this make sense?
No, makes little to no sense. Consider that we are talking about the speed of light speed and what that means per being slow. What happens at the output is what happens at the feedback input and loop. It's a loop that is very very fast.
 

Mnyb

Major Contributor
Forum Donor
Joined
Aug 14, 2019
Messages
2,638
Likes
3,599
Location
Sweden, Västerås
That’s now how you should interpret this time and frequency is a measure of the same thing .
Feedback reach its end when the the lag is such that the signals are -180 degrees apart and it starts a feedback loop .
 

Doodski

Grand Contributor
Forum Donor
Joined
Dec 9, 2019
Messages
20,745
Likes
20,757
Location
Canada
Yes, excessive negative feedback can become destructive when it tries to pass a sharp impulse beyond the normal headroom of the system, and then the latency detracts from the recording. This is especially obvious when comparing phono stages with a record that has a pop/tick - which can linger much longer audibly versus quickly dissipating such that less attention is drawn to it.
Does the feedback have latency under more normal parameters?
 

antennaguru

Senior Member
Joined
Jun 16, 2021
Messages
391
Likes
416
Location
USA
Does the feedback have latency under more normal parameters?
I cannot say I have heard it stand out so clearly as with the phono stage example I offered, but that does not mean it doesn't have subtle effects otherwise which might affect the listening experience - even though it makes the test equipment happy. The phono stage example is rather extreme as there is a real lot of gain and often not so much headroom.
 
Last edited:

sarumbear

Master Contributor
Forum Donor
Joined
Aug 15, 2020
Messages
7,604
Likes
7,313
Location
UK
If heard it assert the the efficacy of negative feedback, and validity of distortion measurements, is really only effective in the case of steady-state signals. Is this true?

It seems to me that there is some latency in a feedback loop, so on the face of it, it sounds reasonable that the feedback might be too slow in case of complex music material. Could this be true? Also most seems it would be true if feedback took milliseconds to arrive but not if it took nanoseconds. I even heard is assert than on account of too-slow response, the feedback in case of complex forms is trying to correct as situation that is already in the past tense. :oops:

Does any of this make sense?
Feedback is part of an amplifier. If feedback is not capable of handling "complex music signals" why do you think the amplifier itself can? Either an amplifier is capable to handle music or not. You are worried with concept that you have no idea of.
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,834
Likes
16,496
Location
Monument, CO
Feedback working with "only steady-state signals" is clearly untrue or virtually none of your devices, from audio to cell phones, would work. Feedback works on all signals (and combinations of signals) within the bandwidth of the feedback loop, which must necessarily significantly exceed the bandwidth specifications of the product. Consider that, if you want low distortion at 20 kHz, then feedback must be effective at 20 kHz without any "latency" problems. That means the feedback loop typically has bandwidth well above (beyond) the audio bandwidth, so works on any signal no matter how complex.

Problems with ticks and pops in a phono stage are usually a problem with overload and input saturation. Feedback may play a part but is not the root cause IME.

Transient intermodulation distortion, TIM, was a problem decades ago (ca. 1970's) as large amounts of feedback were applied without sufficient gain-bandwidth and slew control. It has been solved for about the same amount of time but is still used to prey on consumers.

If the amplifier meets its bandwidth spec then feedback has sufficient gain-bandwidth to do its job. It does not care if it is a 100 Hz sine wave or a complex signal like music or pink noise. The period of a 1 kHz signal is 1 ms; the period of a 20 kHz signal is 0.05 ms (50 us). Loop bandwidth is typically several octaves to a decade greater than the target signal bandwidth so latency is not a problem. If latency was a problem, there would be stability problems with the amplifier. Music signals are normally band-limited, even more so these days since ADCs and DACs require anti-aliasing and anti-imaging filters, so "complex music signals" do not have greater bandwidth than any other audio signal.

EMI/RFI can be an issue but again any competent design handles that just fine. You do not really want your audio system to have too much bandwidth; it greatly increases noise and could amplify undesired signals (like noise from light ballasts and AM radio). Too much can be as disastrous or even more so than too little when it comes to bandwidth.

IME/IMO - Don
 
Last edited:
OP
Gorgonzola

Gorgonzola

Major Contributor
Forum Donor
Joined
Jan 27, 2021
Messages
1,017
Likes
1,398
Location
Southern Ontario
If the amplifier meets its bandwidth spec then feedback has sufficient gain-bandwidth to do its job. It does not care if it is a 100 Hz sine wave or a complex signal like music or pink noise. The period of a 1 kHz signal is 1 ms; the period of a 20 kHz signal is 0.05 ms (50 us). Loop bandwidth is typically several octaves to a decade greater than the target signal bandwidth so latency is not a problem. If latency was a problem, there would be stability problems with the amplifier. Music signals are normally band-limited, even more so these days since ADCs and DACs require anti-aliasing and anti-imaging filters, so "complex music signals" do not have greater bandwidth than any other audio signal.
..
IME/IMO - Don
Whatever else, your comment about 50 uS to meet the requirement for 20 kH, is reassuring that feedback is "facs enough". Of course 50 microseconds is 50,000 nanoseconds and the feedback loop is surely faster than that
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Consider that the vast majority of music has gone though many feedback loops (opamps always use feedback and are in almost every device) from the moment it came from a microphone or 'pick-up'.
And yet only when feedback is avoided in the playback chain the sound really improves or is it that it does degrade even further ?
In any case all the feedback that has come before it hit the playback chain does not seem to have done much damage going of the impressions from people that prefer no overall feedback designs (that usually have lots of local feedback).
 

Mnyb

Major Contributor
Forum Donor
Joined
Aug 14, 2019
Messages
2,638
Likes
3,599
Location
Sweden, Västerås
Feedback working with "only steady-state signals" is clearly untrue or virtually none of your devices, from audio to cell phones, would work. Feedback works on all signals (and combinations of signals) within the bandwidth of the feedback loop, which must necessarily significantly exceed the bandwidth specifications of the product. Consider that, if you want low distortion at 20 kHz, then feedback must be effective at 20 kHz without any "latency" problems. That means the feedback loop typically has bandwidth well the audio bandwidth, so works on any signal no matter how complex.

Problems with ticks and pops in a phono stage are usually a problem with overload and input saturation. Feedback may play a part but is not the root cause IME.

Transient intermodulation distortion, TIM, was a problem decades ago (ca. 1970's) as large amounts of feedback were applied without sufficient gain-bandwidth and slew control. It has been solved for about the same amount of time but is still used to prey on consumers.

If the amplifier meets its bandwidth spec then feedback has sufficient gain-bandwidth to do its job. It does not care if it is a 100 Hz sine wave or a complex signal like music or pink noise. The period of a 1 kHz signal is 1 ms; the period of a 20 kHz signal is 0.05 ms (50 us). Loop bandwidth is typically several octaves to a decade greater than the target signal bandwidth so latency is not a problem. If latency was a problem, there would be stability problems with the amplifier. Music signals are normally band-limited, even more so these days since ADCs and DACs require anti-aliasing and anti-imaging filters, so "complex music signals" do not have greater bandwidth than any other audio signal.

EMI/RFI can be an issue but again any competent design handles that just fine. You do not really want your audio system to have too much bandwidth; it greatly increases noise and could amplify undesired signals (like noise from light ballasts and AM radio). Too much can be as disastrous or even more so than too little when it comes to bandwidth.

IME/IMO - Don

You explained it so well . It’s bandwidth that’s it .
 

Mnyb

Major Contributor
Forum Donor
Joined
Aug 14, 2019
Messages
2,638
Likes
3,599
Location
Sweden, Västerås
It was 30 years since I did any bode or nyqvist plots , so I really can’t explain the theory that well.
DonH56 did it very well

I never designed amps but feedback loops are used in many industrial control applications .

I still do step response test on for example electrical motor drive systems that have a PID controller combined with filters , so we practically find the limit where it verges on oscillation and then dial it back a bit :) the mantra for industrial processes is to not tune the controllers harder than needed for the application, you may invite instability later on when the machinery ages .
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
... If the amplifier meets its bandwidth spec then feedback has sufficient gain-bandwidth to do its job. It does not care if it is a 100 Hz sine wave or a complex signal like music or pink noise. The period of a 1 kHz signal is 1 ms; the period of a 20 kHz signal is 0.05 ms (50 us). Loop bandwidth is typically several octaves to a decade greater than the target signal bandwidth so latency is not a problem. ...
This makes me think of a related question: a typical opamp's raw or open loop response has a constant gain * bandwidth: output drops something like 20 dB per frequency decade. Indeed, -20 dB = 1/10, so when frequency increases 10x, gain drops 10x, which is why gain * bandwidth is constant.

Human hearing is roughly 3 decades: 20 to 20,000 is a ratio of 1000:1. Does that mean the opamp's raw / open loop gain drops about 60 dB over this range? If so, the op amp's open loop output (and thus the negative feedback signal) is heavily bass or low frequency dominated. Subtracting it from the input (or inverting it and combing it with input) effectively applies the reverse gain/frequency curve, squashing the low frequencies at 60 dB / decade which flattens the opamp's closed loop frequency response. If so, that's why negative feedback flattens frequency response or extends bandwidth, by attenuating the low frequencies more than the high frequencies. It's similar to RIAA eq on LPs, or CD redbook emphasis.

If that is right, I have another question. But before proceeding I first want to confirm whether that's right and correct any errors.
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,834
Likes
16,496
Location
Monument, CO
This makes me think of a related question: a typical opamp's raw or open loop response has a constant gain * bandwidth: output drops something like 20 dB per frequency decade. Indeed, -20 dB = 1/10, so when frequency increases 10x, gain drops 10x, which is why gain * bandwidth is constant.

Human hearing is roughly 3 decades: 20 to 20,000 is a ratio of 1000:1. Does that mean the opamp's raw / open loop gain drops about 60 dB over this range? If so, the op amp's open loop output (and thus the negative feedback signal) is heavily bass or low frequency dominated. Subtracting it from the input (or inverting it and combing it with input) effectively applies the reverse gain/frequency curve, squashing the low frequencies at 60 dB / decade which flattens the opamp's closed loop frequency response. If so, that's why negative feedback flattens frequency response or extends bandwidth, by attenuating the low frequencies more than the high frequencies. It's similar to RIAA eq on LPs, or CD redbook emphasis.

If that is right, I have another question. But before proceeding I first want to confirm whether that's right and correct any errors.
Not quite, it is not equalization but compensation. Rather than thinking of it "squashing" low frequencies like an RIAA equalization curve, think of it as more correction being available to suppress errors (distortion). As frequency increases, open-loop gain decreases, true, so less compensation (error correction) is applied and distortion rises. The lower the closed-loop gain, the more open-loop gain is available to correct the output. Note that it is a correction; you can start with a pretty linear circuit (not necessarily an op-amp) that may require less feedback (correction) to attain your target distortion. The signal passes through (amplified by your choice of gain) and the error is reduced by the loop gain.

Simplistically, for a basic inverting amplifier with open-loop gain Aol, feedback resistor RF, and input resistor RIN:

Vout = -Vin * (Aol*RF) / (RF+RIN+Aol*RIN) ~ -Vin*RF/RIN when Aol >> RF+RIN

Now if the open-loop gain is much, much larger than (RF+RIN), the equation simplifies to Vout = -Vin*RF/RIN so the closed-loop gain is RF/RIN. Audio op amps often have 120~140 dB or more of open-loop gain at low frequency and gain stages may be around 12~20 dB so that's a lot of correction (albeit falling at higher frequency) on top of the circuit's intrinsic (open-loop) linearity. Similar equations can be applied to distortion from the op-amp to show it is also reduce by the loop-gain, the difference between the open-loop and closed-loop gain (the "extra" gain available to compensate distortion and other errors). The signal is not "squashed", but errors are, by the available loop gain -- which does decrease with frequency. The signal passes through, amplified by the desired gain, with the difference (error) between input and output corrected by the loop gain.

This is hard without pictures -- I keep wanting to scribble on a notepad or white (or black) board. Here are some references from a quick search:


The first looks best as a starting point after a quick glance.

HTH - Don
 
Last edited:

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
Not quite, it is not equalization but compensation. Rather than thinking of it "squashing" low frequencies like an RIAA equalization curve, think of it as more correction being available to suppress errors (distortion). As frequency increases, open-loop gain decreases, true, so less compensation (error correction) is applied and distortion rises. ...
Sure, the essential operation is "correction", though the downward tilted slope of the raw open loop response implies that equalization results from the correction, which flattens frequency response. Isn't that the reason negative feedback increases bandwidth? But that's not the only benefit of negative feedback. From what I read, the correction also offsets non-linearities in the transfer function, which reduces distortion.

My next question pertains to the fact that the sloped open loop frequency response means there is more correction at low frequencies than high frequencies. But that's the opposite of what we want perceptually. Perceptually we want the most correction around 1000 to 5000 Hz where our hearing is most sensitive to noise & distortion.

So why not apply pre-emphasis and post-emphasis to the gain-feedback loop in amplifiers? Apply a linear attenuation from, say, 1 kHz downward before the gain-feedback loop. Then apply the reverse in the final gain stage. That would focus the noise & distortion benefits of negative feedback into the frequency range where perceptually it makes the most difference. Sort of like frequency shaping dither, one could frequency shape the negative feedback.
This is hard without pictures -- I keep wanting to scribble on a notepad or white (or black) board. Here are some references from a quick search:
...
I've seen a lot of those references and others, reading about negative feedback off and on for years, every time I thought I understood it I realized I really didn't. I've been reading more about it lately to try to understand it.

PS: here's the best explanation I've found so far: https://www.allaboutcircuits.com/te...t-1-general-structure-and-essential-concepts/
I hope any others who are interested also find it useful.
 
Last edited:

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,834
Likes
16,496
Location
Monument, CO
Sure, the essential operation is "correction", though the downward tilted slope of the raw open loop response implies that equalization results from the correction, which flattens frequency response. Isn't that the reason negative feedback increases bandwidth? But that's not the only benefit of negative feedback. From what I read, the correction also offsets non-linearities in the transfer function, which reduces distortion.

My next question pertains to the fact that the sloped open loop frequency response means there is more correction at low frequencies than high frequencies. But that's the opposite of what we want perceptually. Perceptually we want the most correction around 1000 to 5000 Hz where our hearing is most sensitive to noise & distortion.

So why not apply pre-emphasis and post-emphasis to the gain-feedback loop in amplifiers? Apply a linear attenuation from, say, 1 kHz downward before the gain-feedback loop. Then apply the reverse in the final gain stage. That would focus the noise & distortion benefits of negative feedback into the frequency range where perceptually it makes the most difference. Sort of like frequency shaping dither, one could frequency shape the negative feedback.

I've seen a lot of those references and others, reading about negative feedback off and on for years, every time I thought I understood it I realized I really didn't. I've been reading more about it lately to try to understand it.
Answering this is probably too much to attempt, at least for me, via an Internet discussion. I suggest following up on the references. Feedback corrects "all" errors including nonlinear transfer function, offset, some thermal issues, etc. Greater bandwidth is the result of moving down the gain curve; as closed-loop gain becomes lower, you move further out (to the right) on the open-loop gain curve. The affect on noise is... complicated. Most noise is uncorrelated and wideband so not really corrected by feedback, though noise gain is reduced by the feedback factor.

You could shape the feedback but there are trades in performance, such as reducing feedback increases distortion, and can increase output noise. Tanstaafl if you like Heinlein. Because the open-loop gain available is fixed by the op-amp's design, you can't really "shift" loop gain to improve the performance in a given band. You might do that to filter signals outside the band of interest, but that does not really apply here. Look at the picture in the first reference showing open and closed loop gain.
 

DVDdoug

Major Contributor
Joined
May 27, 2021
Messages
2,916
Likes
3,831
Negative feedback (corrective feedback) is a wonderful thing! It lowers noise and distortion while flattening frequency response.

It's similar to RIAA eq on LPs, or CD redbook emphasis.

With a regular amplifier more negative feedback means less gain. And the ratio of resistors determines the amount of feedback to control gain. With no resistors and 100% feedback you can make a unity gain buffer. (A buffer is useful for converting a high-impedance low-current source to a lower impedance higher current source, and it's fairly common in audio.)

Capacitive reactance (measured in Ohms) is higher at low frequencies so you can use a capacitor instead of a resistor (or actually a combination of capacitors & resistors) to get less feedback, and therefore more gain, at low frequencies. With the right combination of resistors & capacitors you can make an RIAA amplifier.

Or, with the capacitors & resistors re-arranged you can make a high-pass filter, or all kinds of active filters.

So why not apply pre-emphasis and post-emphasis to the gain-feedback loop in amplifiers? Apply a linear attenuation from, say, 1 kHz downward before the gain-feedback loop. Then apply the reverse in the final gain stage.
There's a possibility of degrading performance if you reduce the open-loop gain before feedback. And if you monkey around with things that introduce phase-shifts, the negative feedback can get "shifted" into positive feedback, and now you've got an oscillator!
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,422
Likes
4,029
Location
Pacific Northwest
...
Capacitive reactance (measured in Ohms) is higher at low frequencies so you can use a capacitor instead of a resistor (or actually a combination of capacitors & resistors) to get less feedback, and therefore more gain, at low frequencies. With the right combination of resistors & capacitors you can make an RIAA amplifier.
...
Sure. My point is that whatever the shape of the opamp's raw open loop response, negative feedback will flatten it. It so happens the shape of that response is -20 dB / decade, so it looks a lot like the RIAA curve. In fact, according to Wikipedia, the RIAA curve is "only" 40 dB from 20 Hz to 20 kHz. That's a lot, but the opamp's response is steeper, changing about 60 dB over the same range!

I realize the purpose of negative feedback is not equalization. EQ just happens due to the steep slope of the opamp's open loop response. It's interesting that it is so very far from flat. Most amps I've seen have distortion rising somewhat with frequency. I suppose this is a contributing factor to that. Due to the opamp's sloped response, negative feedback applies more correction to the bass, less to the treble.
 
Top Bottom